Structure-Based Neural Network Protein-Carbohydrate Interaction Predictions at the Residue Level

Area of Science:

• Computational biology
• Structural biology
• Machine learning

Background:

• Carbohydrate-protein interactions are crucial for cellular processes like recognition and differentiation.
• Predicting these interactions is challenging due to a lack of reliable computational tools.
• Accurate prediction of carbohydrate-binding sites is essential for understanding these vital biological functions.

Approach:

• Developed two deep learning models: CAPSIF:V (3D-UNet) and CAPSIF:G (equivariant graph neural network).
• Evaluated model performance using Dice scores and Matthews correlation coefficients (MCCs).
• Tested CAPSIF:V on both experimentally determined and AlphaFold2-predicted protein structures.

Key Points:

• Both CAPSIF models outperform existing methods for carbohydrate-binding site prediction.
• CAPSIF:V achieved higher accuracy (Dice score 0.597, MCC 0.599) compared to CAPSIF:G (Dice score 0.543, MCC 0.538).
• CAPSIF:V shows consistent performance across experimental and predicted protein structures.

Conclusions:

• CAPSIF models provide a reliable computational approach for identifying carbohydrate-binding sites on proteins.
• CAPSIF:V is a promising tool for structural biology and drug discovery applications.
• The models can be integrated with docking tools like GlycanDock for predicting protein-carbohydrate structures.